Project management edition
A Bloomberg investigation shows what happens when no one questions an AI recommendation anymore, Atlassian gives Rovo Chat a team memory, and a quarter of parliamentary questions now partly come from a chatbot.
When no one questions the AI anymore: the lesson from the Palantir investigation
A system that cut work from hours to minutes was trusted too much. You'll recognise that pattern in your own reporting too.
On 18 September, Bloomberg published an investigation into a deadly air strike on a school in Minab, Iran, on 28 February. The strike killed 123 children. The investigation found that excessive trust in Palantir's Maven Smart System played a role. That system uses Anthropic's Claude to rank targets in a semi-automated way. Military staff at US Central Command assumed Maven would flag outdated or conflicting intelligence. That didn't happen. Work that normally took hours was done in a few minutes with the system. Palantir says there's no evidence the software malfunctioned, and that the company isn't responsible for the underlying data. It did add new checks meant to catch conflicting intelligence. This isn't project news, and the scale can't be compared to your work. But the mechanism is the same as in your own reporting: a system delivers an outcome faster than a person can check it, so the checking stops. This week, state clearly, for every AI-generated report, who checks it and against what. Maven ranks targets based on incoming intelligence, but it doesn't assess whether that intelligence is old or conflicting: it only sorts what it receives. As long as the list looks complete and orderly, nothing seems to be missing. The speed gain makes this worse, because there's no longer a moment between input and decision where someone naturally looks back to check. The same thing happens with a status report from Jira, Planner or Monday. The tool reads statuses, dates and assignments and turns them into smooth text, without knowing whether those statuses are correct. It's tempting to scrap AI reporting altogether, but that's the wrong lesson: manual work also produces errors, just more slowly. The real trade-off lies in the checkpoint. Any time you save gets taken from somewhere else, usually from the moment a person compares the outcome with their own understanding. So agree, per type of report, who reviews it and against what. For a weekly update, a quick look from the project lead is enough. For a steering group document or a decision paper, you want a second pair of eyes that didn't enter the data themselves.
Background for subscribersThis edition shows how speed and trust reinforce each other: the Maven system, Rovo Chat, Facilitator and Monday's AI columns all work faster than a person can check. So agree, per report or tool, who checks the outcome and against what, before you use it in a decision.
Atlassian gives Rovo Chat a memory and shareable workflows
On 17 September, Atlassian introduced new features in Rovo Chat. You can now see and manage which memory settings are switched on. You can bring agents into a conversation with an @-mention, share chats with colleagues, and create 'custom skills': recurring work saved as a reusable workflow, so you don't have to explain it every time. The update builds on the Teamwork Graph, which tracks how work moves between Jira, Confluence and Jira Service Management. For project leads, the benefit is in handover: the context of a sprint or release plan stays findable when someone takes over. This week, turn your regular weekly status update into one custom skill and have a colleague run it. A custom skill is a saved instruction plus the place the data comes from. You explain once what your weekly update should contain and which boards or pages it comes from. After that, everyone calls up the same skill and gets the same structure. Team memory works like a notepad the chat keeps nearby: agreements and preferences you gave before don't need to be typed again. A memory you can't see is a risk. So check what's being remembered before you share sensitive project information in a chat colleagues can open. And keep in mind that a shareable chat can also spread a half-formed decision. Record decisions in the project, not in the chat history.
Background for subscribersThis is a taster.
Subscribers read the whole Project management edition: every story, each with its background in plain language.
Subscribe →Not ready for that? Read the free letter first →
AI and tech, made simple.
This is the Project management edition. Subscribers get the background to every story, in plain language.
Subscribe
inklaretaal · Amsterdam · © 2026